AgentSkillsCN

bdd-orchestrating

面向功能级别的 BDD 编排器。当用户说“添加”、“实现”、“构建”、“功能”、“迭代”、“故事”、“阶段”、“恢复”、“继续”,或引用票证/迭代/故事时使用。当工作涉及 3 个以上文件的新状态/流程,或用户运行 /bdd 时也使用。不要用于修复 bug、拼写错误、配置更改或 1-2 个文件的任务。

SKILL.md
--- frontmatter
name: bdd-orchestrating
description: BDD orchestrator for feature-level work. Use when user says 'add', 'implement', 'build', 'feature', 'iteration', 'story', 'phase', 'resume', 'continue', or references a ticket/iteration/story. Also use when work touches 3+ files with new state/flows, or when user runs /bdd. Do NOT use for bug fixes, typos, config changes, or 1-2 file tasks.
allowed-tools: '*'

BDD Orchestrator

Behavior-first development for features. Discovery → Scenarios → Implementation.

Iron Law: DEFINE BEHAVIOR BEFORE IMPLEMENTATION

Phase Tracking

Features progress through phases. Track in ticket frontmatter:

yaml
---
type: feature
phase: implement # intake | define-behavior | scenario-gate | decomposition | implement | done
---

Phase meanings:

PhaseWhat happensDetails
intakeContext check, discovery (Phase 0-2)DISCOVERY.md
define-behaviorWriting Given/When/Then (Phase 3)SCENARIOS.md
scenario-gateValidating scenarios (Phase 4)SCENARIOS.md
decompositionTask breakdown (Phase 5)DECOMPOSITION.md
implementOutside-in TDD (Phase 6)TDD.md
doneCleanup, verification (Phase 7)DONE.md

Update phase when:

  • Completing a BDD phase → set next phase
  • Handing off to TDD → set implement
  • All scenarios pass → set done

Resume Logic

When user references a ticket, resume work:

  1. Read ticket → get current phase:
  2. Find progress → first unchecked [ ] in test-definitions
  3. Check context → read last work log entry
  4. Announce resume → "Resuming at [phase]. Last: [log entry]."

Resume by phase:

PhaseResume action
intakeStart context check (Phase 0-2)
define-behaviorContinue drafting scenarios
scenario-gateContinue validating scenarios
decompositionContinue task breakdown
implementFind first unchecked scenario, run TDD
doneRun /done and /audit checks

Current Behavior

  1. Detect work level (see SAFEWORD.md "Work Level Detection")
  2. If user references iteration/story/phase from a spec:
    • Check if child ticket exists for that iteration
    • If not → create ticket, run full BDD
    • If yes → resume at current phase
  3. Announce with override hint
  4. If ticket exists: Read phase, resume at appropriate point
  5. Artifact-first rule: Before doing work, create/verify the phase artifact:
    • Phase 0-2 → ticket at .safeword-project/tickets/{id}-{slug}/ticket.md
    • Phase 3 → test-definitions at .safeword-project/tickets/{id}-{slug}/test-definitions.md
    • Phase 5 → task breakdown in ticket
  6. Execute phase using the appropriate phase file
  7. Update phase in ticket when transitioning

Phase Files

Load the appropriate file based on current phase:

PhaseFile
intakeDISCOVERY.md
define-behaviorSCENARIOS.md
scenario-gateSCENARIOS.md
decompositionDECOMPOSITION.md
implementTDD.md
doneDONE.md

For splitting large features, see SPLITTING.md.


Key Takeaways

  • patch/task → TDD directly (RED → GREEN → REFACTOR)
  • feature → full BDD flow (Phases 0-7), track in ticket phase: field
  • Resume → read ticket, find first unchecked scenario, continue
  • Split → check thresholds at Entry, Phase 3, Phase 5; user decides (see SPLITTING.md)
  • Done gate → run /done before marking ticket complete
  • When unsure → default to task, user can /bdd to override